identified by gene profiling

Genetic regulators of myelopoiesis and leukemic signaling identified by gene profiling and linear modelingAnna L.Brown,*,†,‡,1Christopher R.Wilkinson,*,†,‡,§,1Scott R.Waterman,¶Chung H.Kok,*,†,‡Diana G.Salerno,*,†,‡Sonya M.Diakiw,*,†,‡Brenton Reynolds,*,†,‡Hamish S.Scott,ԽԽAnna Tsykin,§,¶Gary F.Glonek,§Gregory J.Goodall,¶,**Patty J.Solomon,§Thomas J.Gonda,††and Richard J.D’Andrea*,†,‡,2*Haematology and Oncology Program,Child Health Research Institute,North Adelaide,South Australia;†TheQueen Elizabeth Hospital,Woodville,South Australia;Departments of‡Paediatrics and**Medicine and§School ofMathematical Sciences,University of Adelaide,South Australia;¶The Division of Human Immunology and HansonInstitute,Institute of Medical and Veterinary Sciences,Adelaide,South Australia;ԽԽThe Genetics and Bioinformatics Division,The Walter and Eliza Hall Institute of Medical Research,Melbourne,Victoria,Australia;and††CancerBiology Program,Centre for Immunology and Cancer Research,Princess Alexandra Hospital,Woolloongabba,Queensland,AustraliaAbstract:Mechanisms controlling the balance be-tween proliferation and self-renewal versus growth suppression and differentiation during normal and leukemic myelopoiesis are not understood.We have used the bi-potent FDB1myeloid cell line model,which is responsive to myelopoietic cyto-kines and activated mutants of the granulocyte macrophage-colony stimulating factor(GM-CSF) receptor,having differential signaling and leuke-mogenic activity.This model is suited to large-scale gene-profiling,and we have used a factorial time-course design to generate a substantial and power-ful data set.Linear modeling was used to identify gene-expression changes associated with continued proliferation,differentiation,or leukemic receptor signaling.We focused on the changing transcrip-tion factor profile,defined a set of novel genes with potential to regulate myeloid growth and differen-tiation,and demonstrated that the FDB1cell line model is responsive to forced expression of on-cogenes identified in this study.We also identi-fied gene-expression changes associated specifi-cally with the leukemic GM-CSF receptor mutant, V449E.Signaling from this receptor mutant down-regulates CCAAT/enhancer-binding protein␣(C/EBP␣)target genes and generates changes characteristic of a specific acute myeloid leukemia signature,defined previously by gene-expression profiling and associated with C/EBP␣mutations.J. Leukoc.Biol.80:433–447;2006.Key Words:myeloid⅐transcription factor⅐myeloid leukemia ⅐microarray⅐gene expressionINTRODUCTIONA detailed understanding of the molecular regulation of my-elopoiesis is critical for developing new approaches to hema-tological therapy and for diagnosis and treatment of myeloid leukemia.A number of hematopoietic growth factors(HGFs), or cytokines,play a key role in regulating the myeloid lineage. Binding of the HGF ligand to their receptors results in activa-tion of intracellular kinase activity and induction of multiple signaling pathways.This,in turn,mediates changes in cell behavior,ultimately through changes in gene expression asso-ciated with proliferation,survival,self-renewal,and differen-tiation.This response is mediated in part by modulation of the action of a number of lineage-specific transcription factors (TFs),which act by regulating key target genes such as cell cycle regulators,HGF receptors,and mature cell proteins that define particular cell types or lineages[1,2].In addition,these TFs often autoregulate their own promoters and are able to inhibit alternative genetic programs,thus specifying lineage commitment[3–5].In the granulocyte-macrophage(GM)lin-eage,key TFs are the CCAAT/enhancer-binding protein(C/ EBP)family and the ets family member PU.1(for a recent review,see Rosmarin et al.[6]),the ratios of which are critical for cell fate[7].The nature of the links between myeloid HGFs and the action of TFs involved in the cellular response is still largely unclear.To address this important gap in our under-standing of myeloid cell responses,we have been examining how intracellular signals initiated by the myeloid HGFs,GM-colony stimulating factor(CSF),and interleukin(IL)-3impact on transcriptional programs.Given the pivotal roles of HGF signaling in regulating he-matopoiesis,it is not surprising that aberrant HGF receptor activation or activation of pathways downstream of HGF recep-tors is associated with leukemia.A key example of this is activation of fms-related tyrosine kinase3(FLT3),which rep-resents the most commonly mutated gene in acute myeloid 1These authors contributed equally to this work.2Correspondence:Child Health Research Institute,7th Floor,Clarence Rieger Building,72King William Road,North Adelaide,South Australia 5006,Australia.E-mail:Richard.dandrea@.auReceived February23,2006;revised March20,2006;accepted March23, 2006;doi:10.1189/jlb.0206112.0741-5400/06/0080-433©Society for Leukocyte Biology Journal of Leukocyte Biology Volume80,August2006433leukemia(AML)and is constitutively activated by acquired mutation(most commonly,internal tandem duplications)in 30–35%of AML cases[8,9].Aberrant HGF signaling in AML cells can also result from constitutive activation of other re-ceptors[10–12]or signaling molecules[13–15].Autocrine production of GM-CSF and IL-3also occur occasionally in AML[16]and chronic myeloid leukemia[17]and also result in constitutive activation of proliferation and survival pathways. The current model for pathogenesis of AML involves cooper-ation between these mutations and others,which leads to a block in differentiation or acquisition of self-renewal.This is associated with disruption of the normal transcriptional pro-gram,sometimes as a result of lesions involving key myeloid TFs such as C/EBP␣and PU.1[1]but commonly through the interference of translocation-derived fusion proteins[18]. There are overlaps between the effects of these two cooperating classes of mutation,as for example,FLT3activation appears to contribute to a differentiation block in some contexts[19]. With the advent of new,gene-profiling technologies,it has become possible to investigate the molecular basis of myeloid leukemias using approaches that measure global gene-expres-sion changes.Over recent years,there has been a large mass of data generated with respect to the gene-expression profiles of AML patient samples[20–23],sets of genes downstream of leukemogenic TFs[24,25]and associated with FLT3muta-tions[19,26–28].Such profiling needs to be considered to-gether with the gene-expression patterns available for myeloid cell line models undergoing directed differentiation[29–32] and granulocytic populations at different stages[33].In many of these analyses,gene-expression changes cannot generally be associated,a priori,with a specific cellular process(e.g., mitogenesis,promoting,or blocking differentiation,survival, self-renewal),as in these systems,many processes are occur-ring simultaneously.Dissection of gene-expression changes associated with each cellular outcome requires differential activation of these processes in parallel cell systems,such that gene-expression profiles can be monitored simultaneously. With such a system,a linear modeling approach permits iden-tification of different classes of genes based on such a complex set of conditions.With this in mind,we turned to a bipotential cell line model of myeloid differentiation,which displays dif-ferential responses to GM-CSF,IL-3,and activated GM-CSF receptor mutants.The FDB1cell line is strictly growth factor-dependent for survival;however,cells proliferate continuously in the presence of IL-3with only a minimal amount of spon-taneous differentiation to neutrophils,monocytes,and megakaryocytes.In the presence of GM-CSF,FDB1cells dif-ferentiate synchronously along the neutrophil and monocyte lineages with complete differentiation after5–7days[34].This ability to uncouple mitogenesis and self-renewal from differ-entiation while still using physiological stimuli allows a dis-section of these processes,not readily achieved in primary cells,which undergo simultaneous proliferation and differen-tiation and eventual growth arrest and cell death.To increase the power of this study,we included two activated mutants of the GM-CSF receptor,which have differential leukemogenic activity.These two mutants,FI⌬and V449E,are derived from the common signaling subunit(h␤c)for GM-CSF,IL-3,and IL-5(reviewed in Gonda and D’Andrea[35]and D’Andrea and Gonda[36])and represent two distinct classes of activated receptor[extracellular(EC);transmembrane(TM).The two mutants display overlapping,biochemical responses compared with the GM-CSF and IL-3receptor complexes[37]and most likely,represent alternative receptor configurations[35].In vivo,V449E induces a myeloid leukemia consistent with its ability to support the generation of immature myeloid cell lines in vitro[38],and the EC mutant FI⌬induces cytokine-inde-pendent formation of CFU-GM[colony-forming units-GM]and erythroid progenitor colonies and leads to a myeloproliferative disease in murine models[38,39].It is important that the FDB1cell line also responds differentially to these activated GM-CSF receptor mutants.V449E induces factor-independent proliferation of FDB1cells and is able to block GM-CSF-induced differentiation,properties that mimic the ability of this mutant to induce AML in vivo.FI⌬induces factor-independent GM differentiation,reflective of the signals that give rise to myeloproliferative disorder in vivo[34,40].Thus,FDB1pro-vides a manipulable model with which to study downstream signaling and transcriptional responses from cytokines and activated receptors.Here,we have used this model,time-course gene-expression profiling,and linear modeling to dis-sect the molecular mechanisms underlying the differential activities associated with the myeloid growth factors GM-CSF and IL-3and the activated mutants of the GM-CSF receptor. With a view to understanding the molecular control of myeloid differentiation,we focused our downstream and clustering analysis on identification of candidate regulatory genes encod-ing TFs and TF-associated products.In addition,we have used comparative analysis between our data and other studies fo-cused on C/EBP␣and AML to examine further the mechanism of leukemic receptor signaling.Here,we provide evidence for a mechanism of leukemic induction by activated growth factor receptor mutants,which may involve modulation of C/EBP␣activity.MATERIALS AND METHODSCytokines and antibodiesRecombinant murine(m)IL-3and GM-CSF were produced from baculovirus vectors supplied by Dr.Andrew Hapel(John Curtin School of Medical Re-search,Canberra,Australia).Culture and analysis of FDB1cellsFDB1cells were maintained as described previously[34].Infected pools were maintained as described plus1␮g/mL Puromycin(Sigma Chemical Co.,St. Louis,MO).Receptor expression(FI⌬or V449E)was confirmed by staining with a murine anti-FLAG TM antibody(Sigma Chemical Co.)followed by an anti-mousefluorescein isothiocyanate-conjugated antibody(Silenus,Chemicon International,Temecula,CA).Cells were analyzed byflow cytometry using an Epics Elite ESP(Coulter Electronics,UK).If necessary,cells were sorted for expression as described previously[41].For the time courses,cells were washed three times in Iscove’s modified Dulbecco’s medium and cultured in 500BM U/mL mIL-3or mGM-CSF or without growth factor{1BM unit is the amount that gives50%maximal colony formation(CFU-GM)using bone marrow cells;see ref.[42]}.To assess differentiation,cells were centrifuged onto slides and stained with May-Gru¨nwald-Giemsa,and the proportion of differentiated cells was determined microscopically.To assess cell viability, the percentage of cells excluding trypan blue was determined using a hemo-cytometer.434Journal of Leukocyte Biology Volume80,August2006Experimental designThe experiment is arranged as a factorial design studying cell lines and time. For time-course information,each cell population was sampled at six time-points(0,6,12,24,48,and72h),and the time zero(T0)point was used as a reference to obtain relative expression levels following IL-3withdrawal.We also included the parental FDB1cell line maintained in IL-3to establish any baseline differences at T0as a result of expression of the activated receptors. Matched time comparisons were also performed at all six time-points between the two conditions supporting differentiation(GM-CSF and FI⌬)and between the two cell populations expressing activated receptor mutants(FI⌬vs. V449E).Two biological replicate samples were used for each cell population, and a dye-swap replicate was performed for each comparison. Hybridization and data analysisTotal cellular RNA was harvested with Trizol(Invitrogen,Carlsbad,CA)and further purified with the RNeasy RNA purification kit(Qiagen,Valencia,CA). Aliquots of RNA were kept to perform quantitative reverse transcriptase-polymerase chain reaction(QRT-PCR)analysis.cDNA was generated using50␮g total RNA as a template and primed with PolyT(V)N(4.0␮g)and random hexamers(1.0␮g,Amersham,UK).cDNA was labeled with Cy5or Cy3dyes using the Cy-Scribe post-labeling kit(Amersham)as per the instructions. Labeled cDNA was hybridized to microarray slides printed with the CompuGen Mouse OligoLibrary(v2.0,21,99765-mers comprising21,587unique genes) by the Adelaide Microarray Facility(Australia).Slides were scanned using a GenePix3000B scanner(Axon Instruments,Sunnydale,CA),and the Spot package(CSIRO,Australia)was used to identify spots and estimate fore-and background intensities(using a morphological opening background estimator) [43,44].Data analysis was performed in R()using the Limma package of Bioconductor[45,46]and in-house scripts.Arrays were normalized using intensity-dependent spatial normalization and scale normal-ization[47].Spatial loess was also applied to a subset of arrays[48].Linear modeling and F-test-based classifications were performed with the Limma package of bioconductor[45].We estimated,effects for baseline(T0)differ-ences between cell lines change over time in FI⌬(0,72h)and sample by time-interaction parameters between FI⌬and V449E and between FI⌬and GM-CSF.To examine the distribution of differentially expressed genes for a given comparison of interest,we constructed a volcano plot,in which we plotted log2(fold change)on the x-axis and the–log10[false discovery rate (FDR)-adjusted P value]on the y-axis.This generates a volcano-like shape,in which genes at low fold changes are typically not differentially expressed(P values close to1;–log(p)approaching zero),and those with strong evidence for differential expression typically have high fold-change values.Gene ontology (GO)over-representation analysis was performed using GO-STAT using the CompuGen Library as the comparison set and applying a FDR P value adjustment[49].Promoter analysis was performed using the CUREOS data-base(.au)using Transfac matrix M00770(V$CEBP_Q3, www.biobase.de)and requiring mouse and human homology.A␹2test was used to test for over-representation of consensus-binding sites.Additional information may be found under Gene Expression Omnibus(GEO)Accession GSE3333.We have followed the guidelines set out by the Microarray Gene Expression Data Society(/miame).Clustering of TFsGO terms were obtained via the SOURCE website(, UniGene build139)or via direct query of the National Center for Biotechnol-ogy Information gene database.To increase sensitivity further to potential TFs, we used homologene to extracted GO terms for human homologs.We selected all genes,which were children of the terms transcription(GO:0006350), nucleoplasm(GO:0005654),DNA binding(GO:003677),or transcription reg-ulator activity(GO:0030258)or contained the terms“transcription,”“histone,”or“chromatin”in their GenBank Description.A total of2338genes was selected as transcription factor-associated.This was reduced down to340after filtering out genes with nonsignificant F-test values(1ϫ10Ϫ5).Clustering was performed using the Diana algorithm(R cluster package).We evaluated the distance between two profiles using a measure based on that of Bar-Joseph[50] for comparing two time profiles.Briefly,for a gene i,a smoothed spline curve C wasfitted to each of thefive time-course and matched time data sets.The distance between two genes(i and j)for time course p was calculated as d ijpϭ͐t1t2[Cip(t)ϪC jp(t)]2dt/(t2Ϫt1)ϭ͐t1t2[Cijp]2dt/(t2Ϫt1),and the total distance was then calculated as D ijϭͱ¥pϭ1..5d ijpϩ(M V499EϪIL32.The dendrogram was then cut at a height of1.25to define14clusters.Full annotation information of genes in each cluster is listed in Supplementary Table1.QRT-PCRAliquots of RNA prepared for the microarray analysis were subjected to real-time RT-PCR analysis.For this,RNA was treated with DNase(Ambion, Austin,TX),reverse-transcribed with Oligo-dT(Ambion)using Omniscript RT (Qiagen).The sequences of the oligonucleotides used for PCR are listed in Supplementary Table2.Gene-specific PCR reactions were performed for38 cycles using Amplitaq Gold(Perkin Elmer,Wellesley,MA)and recommended conditions.SYBR green(10ϫ;Molecular Probes,Eugene,OR)was added to afinal concentration of0.6ϫper reaction,which was performed on the Rotor-Gene3000and related software used for data collection and to deter-mine mean expression values relative to Cyclophilin A(Corbett Research, Version5.0).Amplification products were analyzed by melt curve and resolved on2%agarose to confirm specificity.Analysis was performed using a mixed-effects linear model to estimate T0differences between each cell line and mean changes within cell lines over time,treating PCR run as a blocking variable. Retroviral transduction and functional analysis in the FDB1cell lineFor functional analysis,the full-length c-Myb cDNA was cloned into the murine stem cell virus(MSCV)-internal ribosome entry site(IRES)-green fluorescence protein(GFP)retroviral vector and FDB1cells infected by cocultivation as described previously[40].Following infection with virus-producing cells,GFPϩFDB1cells were isolated byflow cytometry and expanded in mIL-3.For testing,cells were withdrawn from IL-3and monitored for morphological differentiation in the presence and absence of GM-CSF for 5days,as described previously[40].Changes in cell morphology were re-corded following cytocentrifugation of cells onto a glass slide and staining with May-Gru¨nwald-Giemsa.Association of the V449E gene set with leukemia subsets identified by gene profilingWe downloaded the(MAS5)normalized data used by Valk et al.[51]from GEO (GSE1159-21979).The normalized data were loaded into R for analysis with the Limma package,and the patients were divided into the16clusters(groups) as defined in ref.[51].For each cluster,we compared expression in patients against normal controls,and for each gene obtained,FDR adjusted P value. We then mapped the V449E-associated gene set(see Table4)to the probes on the HG-U133A GeneChip.After cross-species mapping,the V449E-associated data set reduced from44probes to30.Then,for each of the16clusters identified in this study,we looked for association between our V449E gene set and genes differentially expressed in this cluster.We ranked all genes on the array on the basis of FDR-adjusted P values and then examined the distribu-tion of ranks of the30Affymetrix probes homologous to our V449E gene set. Significance was assessed using a Wilcoxon rank sum test as implemented in R.RESULTSA model system of myeloid cell growthand differentiationWe have previously outlined the use of retroviral infection to generate FDB1cell populations expressing the activated GM-CSF receptor mutants FI⌬and V449E[40].A similar level of expression of the hßc mutant in each population was confirmed by staining for the FLAG epitope fused to the N terminus of the mutant␤subunit(Fig.1A).For each condition,changes in morphology,growth,and viability were quantitated(see Fig.1, B and C).The FDB1cells behaved as described previously[34,Brown et al.Regulators of myeloid differentiation and leukemia43540],and complete differentiation to granulocytes and macro-phages (approximately equal ratio)occurred over 5days in GM-CSF and in response to the FI ⌬signal.The switch from IL-3to V449E signaling did not result in any discernible change to morphology,growth,or viability of the FDB1cells.Experimental design and data generationFor dissection of signaling pathways,we performed a time-course study using FDB1cells switched from a continuous growth signal (IL-3)to signaling via the leukemic receptor,V449E,or to conditions permitting synchronous monocytic and granulocytic differentiation (GM-CSF or FI ⌬).This allowed a parallel time-course comparison to reveal gene-expression changes associated with the switch to alternative differentia-tion-inducing signals (GM-CSF or FI ⌬)or signaling via the leukemia-inducing V449E mutant.In addition,we performed comparisons between cell populations at matched time-points to reveal differences in gene expression between cells under-going differentiation or proliferation/self-renewal.The overall experimental design,incorporating matched-time andtime-Fig.1.FDB1experimental system.(A)Cell surface expression of the hßc-activated mutants on FDB1cells as determined by flow cytometry.Transduced cells (solid peaks)were stained with an anti-FLAG TM monoclonal antibody and compared with identically stained,untransduced cells (open peaks).(B)Photomicro-graphs of cells induced to self-renew or differentiate over 5days in the presence of the indicated growth factors or through the activity of the hßc-activated mutants.Original magnification,400ϫ.(C)Differential counts performed with the cell populations used for RNA preparation.Black,Blast cells;light gray,granulocytes;dark gray,monocytes;white,intermediately differentiated cells.436Journal of Leukocyte Biology Volume 80,August 2006course comparisons (along with dye swaps and biological rep-licates),is shown schematically in Figure 2and was based on admissible design criteria [52].A major strength of this design is that it allows estimation of genes with differential expression in the initial cell populations (i.e.,T 0differ-ences)and time-related changes within a cell population (main effects).It is important that this design also permits identification of genes with significantly different expression profiles between cell populations over time (interaction ef-fects).Linear modeling to identify differentially expressed genesThe linear modeling approach is particularly well-suited to factorial experiments comparing one parameter (cell popula-tion)against another (time)and is more powerful than the alternative approach of analyzing the data set as a series of smaller experiments with the same total number of slides [45].We used a linear modeling approach that allowed us to com-bine arrays from different cell populations and use an F-test-based approach to select genes with complex expression pro-files of potential biological interest [45,46].This allowed us to identify common gene-expression changes between FDB1cell populations undergoing one or more fates in common as shown in Figure 2.Thus,we classed differentiation-associated genes as those that increased in expression over 72h in the FI ⌬and/or the GM-CSF cell population and for which this change was significantly greater than the change (if any)in the V449E cell population.The criteria used to select proliferation and self-renewal-associated genes were identical,except that we required decreases in gene expression rather than ing an F-test with a FDR-adjusted P value cut-off of 1ϫ10Ϫ5and a minimum change of 1.4-fold,we identified 205differentiation-associated genes and 175proliferation or self-renewal-associated genes.These gene sets are represented in volcano plots in Figure 3,which provide a visual indication of the relative evidence for differential expression of a given gene.Figure 3A shows a clear increase in the number of genes displaying differential expression over time,based on FDR-adjusted P value,in the parental FDB1population shifted to GM-CSF and the FI ⌬population.Although few significant changes are observed at the 6-h time-point,clear groups of differentiation-associated (red)and proliferation-associated (green)genes are identified by this analysis over the 72-h time-course.To examine V449E signaling as a model of leu-kemic-activated cytokine receptor signaling,we compared the gene-expression profiles of the parental FDB1cells and the FI ⌬population (in IL-3)with the V449E cell population under the same ing an F-test with a FDR-adjusted P value cut-off of 1ϫ10Ϫ5and a minimum change of 1.4-fold,we identified 44genes specifically down-regulated by the V449E mutant (shown in purple in Fig.3B).It is interesting that this gene set is strongly differentiation-associated,as indicated by the increased expression of these genes over time in the parental cells responding to GM-CSF and in the FI ⌬population (Fig.3A).This is most likely reflective of the ability of V449E to block differentiation of myeloid cells,a property that is crucial for induction of AML in vivo.After examining profiles of genes with significant gene-expression changes under various conditions,we selected 10genes of interest and performed QRT-PCR to provide valida-tion for the microarray results (Fig.4).For each gene,we then made eight comparisons and compared significance level and direction between microarray and QRT-PCR-based estimates.For the 80comparisons made,we observed 72.5%concor-dance on significance level and direction.In general,microar-ray-based estimates of fold changes were smaller than those from QRT-PCR.Validation of the model system—measuring gene-expression changes and function in the FDB1systemAs discussed above,we identified differentiation-associated genes on the basis of elevated expression in the FI ⌬and/or the GM-CSF population and stable expression in the V449E pop-ulation.Table 1lists all 205differentiation-associated genes together with their fold-changes and FDR P values over 72h in the FI ⌬cell population (changes in GM-CSF were similar and are supplied along with full gene annotation information in Supplementary Table 3).Example gene-expression profiles are shown in Figure 5A .GO analysis of this gene set indicated significant over-representation of terms such as defense re-sponse (GO:0006952,P ϭ2ϫ10Ϫ8),response to wounding (GO:0009611,P ϭ5ϫ10Ϫ6),and chemotaxis (GO:0006935,P ϭ5ϫ10Ϫ5),consistent with up-regulation of genes involved in mature granulocytes and macrophages.Many of the genes identified in this analysis are known to be associated with myeloid differentiation (Table 2),thus providing an important validation of this cell-line model of myeloid differentiation and confirming the hypothesis that genes displaying concordant regulation in response to the FI ⌬mutant and GM-CSF will be of most relevance to GM differentiation in vivo.We also examined genes for which higher levels of expres-sion are associated with proliferation and self-renewal.We identified these on the basis of their maintained expression at 72h in the V449E-expressing FDB1population andtheirFig.2.Experimental design.Parental FDB1cells grown in IL-3were washed and cultured for up to 3days in the presence of GM-CSF.In parallel,FDB1cell populations expressing FI ⌬or V449E were washed and cultured in the absence of growth factor.RNA was harvested from these populations at 0,6,12,24,48,and parisons performed are represented by arrows (double-headed arrow indicates the use of dye-swap comparisons).All com-parisons presented were performed four times (i.e.,two replicate cultures,two dye swaps,total slides 112).Differential cellular outcomes associated with each condition are indicated on the right.P,Proliferation;S,survival;D,differentiation;L,leukemic signaling.Brown et al.Regulators of myeloid differentiation and leukemia437。